Methods applied to investigage the major UVCE that

occured in the TOTAL refiner’s Fluid Catalytic

Cracking Unit at La Mede, France

P. Michaelis, A. Hodin, Jean-Francois Lechaudel, P. Mejean

To cite this version:

P. Michaelis, A. Hodin, Jean-Francois Lechaudel, P. Mejean. Methods applied to investigagethe major UVCE that occured in the TOTAL refiner’s Fluid Catalytic Cracking Unit at LaMede, France. 8. International Symposium Loss Prevention and Safety Promotion in theProcess Industry, Jun 1995, Anvers, France. pp.365-376, 1995. <ineris-00971929>

HAL Id: ineris-00971929

https://hal-ineris.ccsd.cnrs.fr/ineris-00971929

Submitted on 3 Apr 2014

HAL is a multi-disciplinary open accessarchive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come fromteaching and research institutions in France orabroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, estdestinee au depot et a la diffusion de documentsscientifiques de niveau recherche, publies ou non,emanant des etablissements d’enseignement et derecherche francais ou etrangers, des laboratoirespublics ou prives.

brought to you by COREView metadata, citation and similar papers at core.ac.uk

provided by HAL-INERIS

95-35

Methods applied to investigate the major ÜVCE that occured inthe TOTAL refinery' s Fluid Catalytic Cracking Unit at La Mede,France.

P. Michaelis - TOTAL Raffmage Europe, Paris La Defense, FranceA. Hodin - EDF-CLI, Lyon, FranceJ.F. Lechaudel, G. Mavrothalassitis - INERIS Verneuil-en-Halatte, FranceP. Mejean - METRAFLÜ, Lyon, France

1. SÜMMARY OF THE EVENT

On monday November 9, 1992 at 5:20 a.m. a major U. V.C.E, occured in theGas Plant of the TOTAL refinery's Fluid Catalytic Cracking ünit at La Mede,France. The origin was a 25 cm2 break in the 8" by-pass of the absorber strippercolumn cooler; an amount of about 15 tons of LPG and light naphtha was releasedwithin 10 minutes, covering an area of 14000m2 including Gas Plant, cryogenic,propene and Merox units betöre being ignited on the FCC main furnace. Therewere eight people on shift in the unit: 6 died, one was very seriously injured, andone slightly injured. The total loss including loss of production is estimated at600,000,000$. (see figure l)Direct domino effects resulted from positive and negative overpressures, missiles,and ground shock: fractures of pipework, tank fires and power Station fire.Indirect domino effects were the consequence of fire exposure inducing pipingbursting: 5:22 a.m., rupture of the depropanizer head line with consecutiveexplosion and fireball. (see photograph of figure 2) 5:26 a.m., explosion of thedebutanizer head line with fireball. Let us mention also the explosions of a LPGpipeline and a gasoil line. A total of 60 ruptured pipes were observed.

2. METHODOLOGICAL APPROACH

The methodological approach developed by the investigation team comprises4 main steps:2.1 Gathering evidence

The search for Information involved namely specific data sources:control room hard copy and electronically stored records: no deviation ofprocess operating parameters were observed on the 50 measurement recordsanalyzed;video tapes from amateurs and newsmedia: 17 fllms collected covering aperiod starting at 5h22 a.m. (2 minutes after the first explosion) to 12h00;record from a gas detector 3 minutes before the explosion;

8th International Symposium Loss Prevention and Safety Promotion in the ProcessIndustries, Anvers, 6-9juin 1995, p. 365-376

Figure 1. Chronology of Events Q epicentre 0 location ofvictims

- 2 -

Flgure 2. 5:22 a.m. second explosion, rupture of the depropanizer head line.

Figure 3. Gas Plant area after explosion and consecutive fires.

- 3 -

more than 1750 photographs from audit team, Professional photographersand amateurs were collected and expertised;witnesses Interviews;missiles mapping;seismic waves records collected from 21 stations located between 28 km and409 km from the epicentre.

2.2 Defining potential scenariosThe selection of the potential accident scenarios was based upon four

concepts:The determination of the explosive mass and the location of the explosionepicentre extrapolated from the observed damages in both far-fleld and near-field, using different ways of modeling: multi-energy approach, TNTequivalency and spherical expanding flame model;The identification of all potential ignition sources leading to thedetermination of the one being the origin of the ignition of the vapour cloud:the FCC furnace F301 located at 150m from the epicentre;The exhaustive numbering of the ruptured equipment: 60 fractured pipes andbranch connections, no capacity nor pressure vessel ruptured;The identification of the ruptured pipes that might generate the mainexplosion, using 3 +5 criteria.

2.3 Investigating the possible scenarioFinally, 4 Systems remained selected for a more complex investigation:- a soda tank containing accidentally introduced propane,- the 8" by-pass ofthe absorber stripper cooler,- a l" brauch connection (guillotine break) located on the 8" line between

cooler and absorber stripper,- a 3" LPG pipeline.

2.4 Performing the validation ofthe assumptiousTo emphasize the most probable sequence and to perform the validation of the

assumptions herewith associated, 5 different tools were used:2D model Software for dispersion Simulation: gaussian plume model anddense model. The released vapours are heavier than air: (C3, C4 and lightnaphtha); the refinery is located on ground with slop; at the time of theaccident, Pasquill stability class was FQ 5.Wind tunnel Simulation with a model of the refinery site on a 1/15001 scale,using Richardson similarity.Dynamic process modeling of the absorber stripper column for determiningthe time associated with process operating parameters deviation.Simulation in a similar unit in another TOTAL refinery.Use of metallurgical and mechanical data to confirm the determination of theprimary leak.

- 4 -

3. INVESTIGATION INTO THE AMOUNT OF FLAMMABLE PRODÜCTSINVOLVED

Inspection of the accident site is one of the most important steps in gatheringphysical evidence. A detailed examination is made to obtain Information on theapparent origin, the propagation of the explosion and localized damage of theexplosion or fire. In the case of explosion damage analyses, the main goal is theevaluation of the amount involved, because hereafter it is a very importantcriterion to classify possible scenarios. Explosion damage is due to the high anddynamic pressures generated by the sudden release of energy. However, thedamage analyses include consideration of the following important items:1. flammable vapour cloud confinement in the Installation2. explosion pressures, propagation rates and energy release3. near-fleld effects4. blast wave or far-field effects.

All these aspects were kept in mind during the damage investigation. Adamage table was set up (see table l). The evaluation of the pressure in the far-field was made by comparison with tabulated blast wave criteria.Related to the near-field, some mechanical caiculations were performed. A mapof the fire damage was established too.

Table lSome damages observedObservation point

Cryogenic unitPropene unitGantry pipe and pipelinesGas plant

Electrical Station 42Electrical Station 40Locker roomControl room

Tanks A38, C24, C25, B14,B56Technical building

Neighbouring habitationsLa MedeCommercial CenterMartigues, Jonquieres andFerneresMartigues, LEP lieu-ditBrise-LamesMartigues SwimmingpoolChäteauneuf-les-Martigues

Effects

mechanical rupturesmechanical ruptures and projectileover balancingbending of a column

all walls breaking downall walls archedbreaking downlarge eardrum rupture

deformation and displacement

inside, structural damages and glassWindows projectionglass Windows breakageabout 50% glass Windows breakagelarge Windows breakagemore than 10% of Windows breakage

several Windows breakage

several bays Windows breakagesbay window breakage

Incidentpressure(kPa)50-10050-10040-5540-707040-8030-5030-5040-905015-25

8-12

5-72-31-21-2

1,5-3

1,5-30,5-1,1

Distance(m)

epicentre areaepicentre areaepicentre area20

35505080

150-170

200

7001000-140021003500-4500

3700

39004000

- 5

One may think thafc the damages behind the refinery fence, in particular atMartigues, were enhenced due to the meteorological conditions at the time of theexplosion. As a first step, the explosion damage analysis allows to determine theepicentre area of the explosion. This area is defined äs the location where thepressure was the highest. This epicentre area was located between the propeneunit and the Gas Plant. It is different from the ignition point and from the originof the leak, because the explosion pressure depends strongly on the congestedparts on the site. (see photograph of figure 3) As a second step, the examinationof the projection of missiles allows to evaluate the pressure at the epicentre area:50 to 200 kPa locally. Then, several methods were used to determine the amountof flammable mixture involved in the explosion: TNT equivalency, multi-energymethod and spherical expanding flame model. For example, the multi-energymethod allows to make some comparison between the congested volume from thedifferent units and the damages observed. The objective was to obtain the bettercurve äs possible to reduce the distance between the curve and the damage points(see figure 4). Finally, the different methods led to good agreement about theassessment of the most probable amount of flammable mixture involved in theexplosion. The best estimation was about 5 tons scale of sizes.

4. SELECTION OF THE FRACTURED PIPES POSSIBLY IMPLICATEDWITH THE PRIMARY LEAK

60 fractures of pipework were observed. In order to identify the ones thatmight generate the main explosion the following methodological approach basedon 3 + 5 criteria was defined. A first selection was made, discarding threefamilies of ruptured pipes: •1. 14 pipes containing products unable to generate an explosion phenomenon

such äs non flammable products (DEA, water, soda) or flammable productswith pressure vapour less than lOhPa at operating temperature (GO, HV fuel,heavy naphtha).

2. 11 pipes being isolated at the time of the explosion and having a smallcapacity.

3. 12 pipes containing flammable products with specific physical conditions:vapour phase and small pressure (1050hPa).For such emission sources we have a mass flow of 2,3 kg/s in vapour phase, anexplosive mass of about 2kg and a LEL concentration up to 45m from thesource with stability class Fl.

For the 23 remaining ruptured pipes, 4 criteria correlated with caiculations andone criterion based upon visual inspection are then applied:1. vapour mass flow must be superior to 10 kg/s;2. the LEL of the released gaseous mixture must reach the furnace;3. 20% of the LEL must be snuffed at the location of the gas detector having

given an alarm 3 minutes betöre the explosion happens;4. the released mass within the explosion limits must be about several tons;

- 6 -

Comparison with observed damages.

1 -» LA MEDE Distance (m)2 -»• MAETIGUES

Figure 4. Multi-energy approach: correlation with observed damagescorresponding to an hemispherical cloud with a 29m radius.

Figure 5. Ignition source: FCC main furnace.

- 7 -

5. The physical conditions of the rupture are not a consequence of the mainexplosion.

18 fractured pipes were consequently elaminated by criteria l to 4. The fiveremaining ruptured pipes were then submitted to criteria 5. Three Systems thennecessitated a more complex analysis:

- 8" by-pass ofthe absorber stripper cooler;-1" branch connection on the 8" line from the absorber stripper cooler;- 3" butane pipeline T09.

5. ANALYSIS OF POSSIBLE LEAKS

Four likely scenarios were considered. The first is related to a leak on theLPG line T09. When the accident occured, it was isolated and füll of liquidbutane. It presents a bulge-shaped breach (200xl20mm). Owing to the diameterof the line and the pressure inside it, the order of magnitude of the explosiveamount of gas would be at the most 80kg, the explosive area extending äs far äs45m. Moreover, witnesses saw the line exploding and metallurgicalinvestigation led to plastic deformation due to heating. It was thus possible todiscard this hypothesis.The second scenario is related to C24 tank in which is usually sent the used sodacoming from the refining of propane, butane and gasoline. When used soda isdrained off, hydrocarbons may be accidentally introduced in the tank. The tankhas burnt the day the accident occured. Modeling was performed related to leaksofgas through vents or broader holes, according to the maximum propane releaserate that might have been drained to the tank, the maximum explosive mass hasbeen assessed to about 1,2 tons, the maximum extension of the explosive cloudbeing 100m. The case of leak without kinetic energy has been considered too, bymeans of a wind-tunnel Simulation äs described below. It was not possible toguess measurable concentrations in the gas plant. All these reasons led to thediscarding of the scenario.The two last scenarios are related to the absorber stripper DA101, the first abouta l" branch connection and the second about a breach (800x200mm) on the by-pass of the EA103 cooler. More complex simulations including wind tunnel anddynamic process modeling were then used. It was thus possible to emphasize thatthe most probable origin of the initial leak was the by-pass äs shown below.

6. IGNITION SOURCE LOCATION

In an unit, the possible ignition sources may be:an electrical spark due to the rupture extra-current of a circuit, generated bysome manoeuvre,an open flame like a burner of a furnace,a mechanical spark,

-8

hot surfaces the temperature of which being greater than auto-ignitiontemperature of the flammable products involved.

In our purpose, the investigation dealt with the electric sources and the F301furnace. In fact, the mechanical spark was not taken into consideration becausethere was no work in progress, nor abnormal Situation in an electrical equipment.The possibility of an ignition on the hot surfaces of the disaster area was not heldbecause the auto-ignition temperature of the gases was greater than theoperating temperature of the pipes.On the other hand, several fmdings were made under the F301 furnace. Theflammable cloud burnt up to the two thirds of the south part of the furnace.Traces of combustion were found on the external box of the third burner at east.More, in the direction of the Gas Plant, several burns, were found on theelectrical cabinets and cables indicating the travel of a slow flame front. Severalinvestigations on the electrical equipment were made in order to justify theexclusion of this kind of equipment.Finally, the ignition on the F301 furnace was found to be the only solution that isconsistent with evidence and with the observed traces of combustion. (seephotograph of figure 5).

7. VALIDATION OF THE ASSUMPTIONS

7.1 Wind tunnel Simulation ofthe dispersion ofthe explosive vapour cloud.The choice of a physical model Simulation was dictated by the difficulty of

reproducing the complexity of flow phenomena on a site such äs La Mede, due tothe lack of available accurate modeis. The experiments were conducted in thediffusion wind tunnel of EDF's Lyon Engineering Center located in the premisesof the Fluid Mechanics and Acoustics Laboratory ÜRA CNRS 263 of the "EcoleCentrale de Lyon". This wind tunnel is of the feedback type with ventingdownstream of the test facility. The test section measures 14 meters in length,3.70 meters in width and the height is variable between 2 and 2.50 meters. Thespeed inside the test section is continuously adjustable from 0-lOm/s. The modelof the La Mede reflnery site on a l/ISO111 scale, represents the whole of the zoneaffected by the explosion and incorporates the potential release points to beinvestigated (see photograph of figure 6). The ground incline is perfectiyreproduced on the model. The physical model study of the dispersion of anexplosive gas cloud heavier than air is performed in Richardson similarity theonly scale parameter of which is represented by the Richardson number

Ri^gPszJ^^0 pa U2

where g is gravitational acceleration,

^/pa

- 9 -

g' is the gravity term, pg and pg., L and U are respectively the speciflc masses ofgas and air, the characfceristic length, and speed scales defined in the case of acontinuous gas release at the break of volume flowrate Qv via

O2 1 iL = (^sandü = (Qvg')3.

g

This definition of the Richardson number, used by Köning-LangIo andSchatzmann (1990) is routinely used for continuous releases in a calmatmosphere, characterized by wing speeds under l m/s. A different definition isused for non-zero wind velocities, where the local flow velocity is used todetermine the characteristic scales (Britter and McQuaid 1988). Validation tests,presented in particular by Köning-LangIo et al (1990) Hall et al (1982) and Y.Riou (1987), have shown good agreement of the wind tunnel results with thoseobtained in situ, mainly at Thorney Island (MC Quaid 1984), Porton (Picknett1978, 1981) for instantaneous gas releases and at Burro (Koopmann et al 1982),Maplin Sands (Puttock) et al 1982) and Thorney Island for continuous gasreleases. Instantaneous concentrations are measured by means of a hydrocarbideanalyzer (Cambustion Ltd) the passband of which is about 500 Hz for a 150 mmsampling probe lenght. The tracer gases used in the wind tunnel are mainlyhydrocarbides whose densities are dose to the considered release conditions. Themain test results for the three scenarios under consideration have shown, despitethe very conservative hypotheses adopted for this Simulation, that the maximumgas leak from storage tank C24 could not be the cause of the accident, since theconcentration levels reached at the ignition point are far lower than the gasinflammability limit (LEL). Regarding the two absorber stripper cooler EA 103scenarios, the tests showed that scenario break at the by-pass bounds the secondone (failure of l" branch connection). The gas concentrations at the ignition point(furnace F 301) only reach the LEL for the first scenario. The Simulation madefor this scenario by east wind led to the results dosest to the evidence. Note inparticular that the gas doud reaches both the gas detector P80-1 and the ignitionsource. The dispersion and transport of the gas doud particularly made itpossible first, to characterize the progression of the explosive mixture äs far ästhe presumed ignition source and, secondly, to evaluate the average time to reachthe ignition point by the LEL at 675 ±110 seconds for 98 simulations, compatiblewith the estimations made elsewhere (see figure 7).

7.2 Dynamic process modeling of the absorber stripperIf one of the breaks is the origin of the accident, it must be demonstrated that,

during the estimated release duration about 10 minutes,1. the break can release a sufficient mass of product2. the break is not inducing variations of the operating parameters, äs oberved

on the hard copy stored records.To verify both criteria a realistic dynamic process modeling of the absorber

stripper column has been performed by the IFP (Institut Francais du Petrole) inorder to determine the time associated with process operating parameters

-10-

Figure 6. Wind tunnel Simulation: 1/150^ scale model of the zone affected by theexplosion.

Figure 7. Wind tunnel Simulation: dispersion and transport of the vapour cloud(8" by-pass case) äs a function oftime,

- 11 -

deviations. This evaluation technique combines physical modeis for quantifyingrelease outflow with transfer functions calibrated by mean of test done on asimilar column in another TOTAL refinery. From that dynamic Simulation it hasbeen demonstrated that for outflow opening areas less than 35cm2, pressurevariations are not detectable before a 30 minutes period; for hole areas less orequal to 25cm2, no noticeable variations is observed for the absorber bottomlevel, for the head vapour flow and for the bottom liquid flow, if the release lastsless than 10 minutes. Taking into account the operating process conditions atthe time of the accident, this dynamic process Simulation has shown that within10 minutes:1. 12 tons of hydrocarbons might have been released through a 25cm2 break

area, the third part of this amount being vaporized.2. Through the l" branch connection 5 tons might have been released, thus

discarding this branch connection scenario.

8. CONCLUSIONS

From the experience gained at La Mede, it was possible to point out thefollowing recommendations to perform an accident investigation:1. It is required for companies to develop incident investigation System

involving top management.The Chairman of the investigation team has to be a manager experienced inoperating production units, to know technology and process of the implicatedkind ofunit and, fmally, has to be aware of safety aspects.

2. An investigation team has to be constituted with internal and external multi-disciplinary experts (fluid mechanics, detonics, mechanics, electricity, •metallurgy, industrial risk analysis, process control...).

3. It is necessary to visit the site äs soon äs possible to identify and secure areaof interest and photograph the maximum of exhibits.

4. It is recommended to associate operators and maintenance crew to theinvestigation, particularly when gathering evidences and trying todemonstrate the feasibility of scenarios.

5. Validation of the assumptions must be performed by using different modelingand Simulation approaches in order to point out the most probable root causeevent.

6. In order to increase the knowledge in learning from accidents, it would beusefui to collect äs precisely äs possible and disseminate all the relevantInformation: lectures, papers, investigation report with non restricteddistribution... .

-12-